Participatory AI for inclusive crop improvement

Area of Science:

• Agricultural Science
• Plant Breeding
• Computational Biology

Background:

• The Global South faces a 'phenotyping bottleneck' in crop breeding due to manual, error-prone methods, hindering climate-resilient variety development.
• Traditional breeding programs often exclude farmers, leading to low adoption of improved crop varieties.

Purpose of the Study:

• To democratize crop breeding using mobile phenotyping technologies for decentralized, high-quality data collection.
• To integrate farmer preferences into variety development using multimodal data, ensuring alignment with local needs.

Main Methods:

• Development and field-testing of novel mobile phenotyping tools in Colombia and Tanzania.
• Analysis of data quality and accuracy against traditional methods and ground truth.
• Application of Human-Centered Design (HCD) to ensure technology usability for breeders.

Main Results:

• Computer vision (CV)-enabled mobile phenotyping significantly reduced scoring variation.
• Achieved high trait accuracies (Pearson Correlation 0.88-0.95) with ground truth data.
• Reduced labor by enabling plot phenotyping in under a minute.

Conclusions:

• Mobile phenotyping allows anyone with a smartphone to collect high-quality data, expanding on-farm phenotyping capabilities.
• Inclusive innovations in mobile phenotyping and AI accelerate trait assessment and farmer involvement in variety selection.
• These advancements facilitate rapid varietal turnover, enhancing food security and cultural relevance of crop varieties under climate change.